ISO; 300D vs 30D and 20D

[PEACHMAN]

I got some grainy photos shooting at ISO 800 this weekend and it has been brought to my attention that the 300D does not produce the same results at these ISO's as the 30D and I assume the 20D(?)...I thought they used the same processor ?? Antbody out there that can streighten me out on this??

[Mikelangelo]

The 300D is a 6 megapixel CMOS procoessor. From what I understand, the 30/20Ds handle higher ISO settings much better. The 20D and 30D DO share the very same CMOS processor, tho'... so it should be virtually identical results with those cameras.

[Steve Parr]

I used to shoot with the 300D but, when I got into concert photography, I needed something that would perform well at higher ISO's (800+). The 20D and 350 were comparable, but I went with the 20D, primarily, because it's physically bigger and fit my hands better.

In my experience, the 300D is downright offensive at an ISO over 400.

If a new body isn't something to consider, you could always go the Neat Image route (noise reduction software). I've had pretty good results with that...

[nitsch]

30D/20D perform better than 300D. I'm with Steve, I don't like the results of the 300D above ISO400, however I quite happily use the 30D at ISO800. The 30D performs identically to the 20D AFAIK however the advantage of the 30D is that you can choose ISO in 1/3 stops - I like this as there are times when you need more than ISO400 but on the 20D you would be forced to leap a whole stop to ISO800 wheras on the 30D ISO500 and 640 are available and if they will suffice then the noise performance is better than pushing it to 800. HTH! :)

[PEACHMAN]

Well , I guess I'll be getting that 20D out a lot more and the 300D a lot less...Thanks for the confermation folks...And 1/3 stops of ISO in the 30D ? I guess that makes more sense than 2X stops...

[peterdoomen]

Yes, the 20D performs quite well in low light and high ISO.
(See for example the results of a show shooting I posted on http://photography-on-the.net/forum/showthread.php?t=133041)

P.

[Tee Why]

No, the 10d/300D use a Digic I processor and have the same IQ.
The 20D/30D uses a diff. sensor and digic II processor. The high ISO noise is a small bit lower.
Here is a noise comparison by dpr of 10D and 20D which should be the same as 300D vs a 30D.
http://www.dpreview.com/reviews/canoneos20d/page20.asp
There is some diff. at ISO 800 and up but I really don't think it's much.

[mbellot]

The 30D performs identically to the 20D AFAIK however the advantage of the 30D is that you can choose ISO in 1/3 stops - I like this as there are times when you need more than ISO400 but on the 20D you would be forced to leap a whole stop to ISO800 wheras on the 30D ISO500 and 640 are available and if they will suffice then the noise performance is better than pushing it to 800. HTH! :)

I've read numerous time that the 1/3 ISOs are "faked" in the 30D and you can get the same effect on the 20D with exposure compensation and some tweaking in PP (ISO500 would be ISO400 with -1/3EC pushed +1/3 in PP, assuming you're shooting RAW).

I've never seen any actual tests/data to back it up, but it makes sense.

[nitsch]

I've read numerous time that the 1/3 ISOs are "faked" in the 30D and you can get the same effect on the 20D with exposure compensation and some tweaking in PP (ISO500 would be ISO400 with -1/3EC pushed +1/3 in PP, assuming you're shooting RAW).

I've never seen any actual tests/data to back it up, but it makes sense.

Very true, but it's a hell of a lot less PP to be able to do it in camera!

You can use the term "faked" if you wish, but what you definately cannot say is that it is just the 1/3 stops in the 30D which are "faked". If you want to use the term "faked" then you must use that term for ALL ISO settings above the native ISO of the sensor on ALL digital cameras. Increasing the ISO in a digital camera is NOT the same as using different film speeds; an ISO400 film is actually twice as sensitive to light as an ISO200 film, however when you switch your digicam from ISO200 to ISO400 your CMOS/CCD has not changed in sensitivity, the signal from the sensor is merely being amplified - you can call this "faking" the ISO value if you wanted I guess, and you are right in saying that there is no reason why you can't do this in the RAW conversion rather than "in camera", but remember this is the same for every digital camera, not just the 30D! HTH! :)

[syntrix]

I thought my 20D had the 1/3 stop option in the Cn menu. Hmmmm.

[adas]

Yes, but only for aperture and shutter, not ISO. :(

[syntrix]

Ahhh, that's right!

[mbellot]

Very true, but it's a hell of a lot less PP to be able to do it in camera!

No argument at all, and if you're shooting jpeg its a big plus.

Its also nice since (I assume) you still have the +/-2 EC that the 20D has already. Though you could get into some interesting numerical twists... Is shooting at ISO500 with +2/3EC more or less noisy than ISO800 (and so on, ad nauseum).

You can use the term "faked" if you wish, but what you definately cannot say is that it is just the 1/3 stops in the 30D which are "faked". If you want to use the term "faked" then you must use that term for ALL ISO settings above the native ISO of the sensor on ALL digital cameras. Increasing the ISO in a digital camera is NOT the same as using different film speeds; an ISO400 film is actually twice as sensitive to light as an ISO200 film, however when you switch your digicam from ISO200 to ISO400 your CMOS/CCD has not changed in sensitivity, the signal from the sensor is merely being amplified - you can call this "faking" the ISO value if you wanted I guess, and you are right in saying that there is no reason why you can't do this in the RAW conversion rather than "in camera", but remember this is the same for every digital camera, not just the 30D! HTH! :)

Well, sort of.

My understanding was that the full stops actually changed analog multipliers (opamps) while the 1/3 stops were handled in the digital domain.

If the 30D actually adjusts the gain on the opamp in 1/3 stop increments then I would consider that "real" ISO. Doing it after the A/D conversion is what I thought was happening and why I called it "fake".

[Titus213]

Peachman - I still use my 300D and don't mind using the higher ISO. I also use Neat Image Pro. The 20D does handle the higher ISO better IMO but the 300D is quite useable. I think the main trick is to get proper exposure as any under exposed image will show some noise. Here's a link to some interesting test results from DPReview. The 300D is a bit long in the tooth at this point so the comparisons are not with current equipment.

http://www.dpreview.com/reviews/canoneos300d/page13.asp

[solinger]

Higher ISO gets grainy when you miss the exposure a little. My 300d is fine at higher ISO but I am not as picky about noise as some. I can try to dig up an 800 ISO shot I took right after I got my 300d that impressed me.

[nitsch]

No argument at all, and if you're shooting jpeg its a big plus.

Its also nice since (I assume) you still have the +/-2 EC that the 20D has already. Though you could get into some interesting numerical twists... Is shooting at ISO500 with +2/3EC more or less noisy than ISO800 (and so on, ad nauseum).

This is a good point, I've done some experimenting in the past with ISO1600 with -1 EC and pushing it in post to 3200 vs shooting at 3200 in the camera. My results were by no means scientific but there were several shots where the pushed 1600 was better (to my eyes) than the 3200 shot. Go figure!


Well, sort of.

My understanding was that the full stops actually changed analog multipliers (opamps) while the 1/3 stops were handled in the digital domain.

If the 30D actually adjusts the gain on the opamp in 1/3 stop increments then I would consider that "real" ISO. Doing it after the A/D conversion is what I thought was happening and why I called it "fake".

OK I see where you are coming from now - so what you're saying is that you do not view analog amplification as "faking" the ISO, right? Makes sense.

My personal view is that the sensor has a native sensitivity of say ISO100 so anything else is merely an amplification of the signal the sensor produces (either before or after the analog / digi conversion). The sensitivity of the sensor itself never changes so anything other than native is "fake".

Does it really matter? I guess not! ;-) Good to hear your view and at least I understand your thinking behind the 1/3 stops now. :)

[solinger]

Here it is. I have gotten much grainier (is that a word?) pictures with my 20d. It just depends on the exposure.

http://i7.photobucket.com/albums/y291/tborquez/August%202006/041705Jackson01.jpg

Hope it helps.

[EDIT] Exif:
Camera Make: Canon
Camera Model: Canon EOS DIGITAL REBEL
Image Date: 2005:04:17 14:18:35
Flash Used: No
Focal Length: 105.0mm
CCD Width: 5.90mm
Exposure Time: 0.0063 s (1/160)
Aperture: f/5.6
ISO equiv: 800
White Balance: Auto
Metering Mode: Matrix

[nitsch]

Here it is. I have gotten much grainier (is that a word?) pictures with my 20d. It just depends on the exposure.

Hope it helps.

Nice shot Sam. You are right about correct exposure being absolutely key.

[mbellot]

This is a good point, I've done some experimenting in the past with ISO1600 with -1 EC and pushing it in post to 3200 vs shooting at 3200 in the camera. My results were by no means scientific but there were several shots where the pushed 1600 was better (to my eyes) than the 3200 shot. Go figure!

Yeah. My problem is I sometimes need to shoot @ ISO3200 with -1EC for shutterspeed needs, and I doubt pushing ISO1600 by two full stops is a good idea.



OK I see where you are coming from now - so what you're saying is that you do not view analog amplification as "faking" the ISO, right? Makes sense.

My personal view is that the sensor has a native sensitivity of say ISO100 so anything else is merely an amplification of the signal the sensor produces (either before or after the analog / digi conversion). The sensitivity of the sensor itself never changes so anything other than native is "fake".

Does it really matter? I guess not! ;-) Good to hear your view and at least I understand your thinking behind the 1/3 stops now. :)

Yup, thats how I look at it. Mainly because the analog signal is a continuous signal so amplification (in theory) has no real "resolution", but once converted to digital form the sensor data has a finite number of discrete values and any mathematical transformation performed on it can have a negative effect on it.

[solinger]

Mainly because the analog signal is a continuous signal so amplification (in theory) has no real "resolution", but once converted to digital form the sensor data has a finite number of discrete values and any mathematical transformation performed on it can have a negative effect on it.

:shock: Wow, that's a mouthful right there...

[cdifoto]

Too much "geek speak", not enough "who cares". ;)

[kram]

I got some grainy photos shooting at ISO 800 this weekend and it has been brought to my attention that the 300D does not produce the same results at these ISO's as the 30D and I assume the 20D(?)...I thought they used the same processor ?? Antbody out there that can streighten me out on this??

Nope - the 300D is definitely not the same at higher ISOs. In fact, I hardly get decent ISO 800 snaps from the Drebel.

Did I tell you my 20D arrives by fedex tomm ;) . Oh yeah, that's of no relevance to the discussion. But if I still have my Drebel, I will get you some comparison shots. But yes, my 20D arrives tomm :).

[Nikolas]

No issues with the 300D and high ISO here.
Just get the exposure right and it should be good.
Also have decent lenses.
I have the hacked version with 3200 iso as well.

800 iso kit lens no noise reduction but used flash
http://home.swiftdsl.com.au/~smeegles/Stuff/Johno.jpg

3200 iso pentacon 135 (with noise reduction)
http://home.swiftdsl.com.au/~smeegles/Stuff/lenstests/car3200iso.jpg

crop
http://home.swiftdsl.com.au/~smeegles/Stuff/lenstests/car3200iso100crop.jpg

800 iso slight noise reduction
http://home.swiftdsl.com.au/~smeegles/Stuff/lenstests/Bonfire/CRW_3307.jpg

800 iso no noise reduction but obviously post processed.
http://home.swiftdsl.com.au/~smeegles/Stuff/Expression3.jpg

[S. William Davis]

I use both the 300D and the 30D and find that at 400 ISO there dosen't seem to be much differents beween the two. I do mosly landscapes and normally will not go over 400. I also shot raw and at 100 iso mostly. For my work there is even to much noise with the 30D over 200 iso unless I am shooting B&W.
Sam

[Jon]

Yup, thats how I look at it. Mainly because the analog signal is a continuous signal so amplification (in theory) has no real "resolution", but once converted to digital form the sensor data has a finite number of discrete values and any mathematical transformation performed on it can have a negative effect on it.It's alwys been my understanding that CMOS and CCD photoreceptors as used in sensors are quantum devices: they don't provide a continuously-variable signal, but rather discrete steps as the raw output. Thus, whether the amplification was hard-wired or handled in software, or both, it'd still be a "digital" amplification. If they truly provided an analog output signal, RAW files should theoretically be able to provide more than 12-bit channels, ands it'd make sense to match the signal to 16-bit TIFF capabilities.

[mbellot]

It's alwys been my understanding that CMOS and CCD photoreceptors as used in sensors are quantum devices: they don't provide a continuously-variable signal, but rather discrete steps as the raw output. Thus, whether the amplification was hard-wired or handled in software, or both, it'd still be a "digital" amplification. If they truly provided an analog output signal, RAW files should theoretically be able to provide more than 12-bit channels, ands it'd make sense to match the signal to 16-bit TIFF capabilities.
The initial conversion from light to electrons is most definitely analog (much like a solar cell). Its quite possible that the sensor includes the A/D as well as other image processing functions (one of the advantages of CMOS is the ability to integrate additional IC functions on the same chip as the sensor) so that what comes off the chip is already digital.

For a brief intro to image sensors of all types you may want to check out this (http://www.shortcourses.com/pixels/sensors/0-sensors.htm) web page.

We are probably stuck at 12 bits because of cost, a high speed 16 bit A/D is quite a bit more expensive than a 12 bit one. I'd also hazard a guess that squeezing useful data into those bits (versus just random noise) is also something of a design challenge in such a tight package.

Designs I've worked on that had > 12bit A/D converters usually segregated the analog section both physically and electrically from the digital section, otherwise switching noise from the digital ICs would couple into the analog section rendering the values in the lowest bits essentially worthless.

[adas]

Designs I've worked on that had > 12bit A/D converters usually segregated the analog section both physically and electrically from the digital section, otherwise switching noise from the digital ICs would couple into the analog section rendering the values in the lowest bits essentially worthless.

I'm willing to believe that there are well isolated sections in that CMOS sensor, with separate +V, -V and GND pins for all sections. So no, the A/D converter caused noise must be minimal. I mean, there are crystal clear 16 bit A/D converters in CD players ever since the mid '80s.
The noise is analog.
Back to the thread, I don't think that the high ISO noise reduced images of the 300D to match the noise levels of the 20D/30D, will soon match the detail level of these too.

[Nikolas]

Back to the thread, I don't think that the high ISO noise reduced images of the 300D to match the noise levels of the 20D/30D, will soon match the detail level of these too.
Must have been the lack of sleep but I don't understand what you mean in this statement.
Can you rephrase it?

[adas]

I mean, reducing noise on the 300D will further reduce detail.

[Nikolas]

I mean, reducing noise on the 300D will further reduce detail.

Really?
Have you seen my 3200 iso example above?
If anything if you apply noise reduction intelligently you can enhance details, and that's not just with the 300D.

[mbellot]

I'm willing to believe that there are well isolated sections in that CMOS sensor, with separate +V, -V and GND pins for all sections. So no, the A/D converter caused noise must be minimal. I mean, there are crystal clear 16 bit A/D converters in CD players ever since the mid '80s.
The noise is analog.

There very well may be good electrical isolation (from a noise standpoint), but physical isolation is also important when you are talking about such small voltages. On an image sensor you don't exactly have acres of real estate to spread things out.

Assuming the A/D is fed an amplified voltage with a full scale value of 3.3 volts (probably lower, but what they hey). that means the least significant bit corresponds to a voltage of 0.00005 volts, or 50 microvolts. Power supply fluctuations can easily be 10 to 50 millivolts on a "good" analog supply, getter better means significant cost increases.

CD Players use D/A (digital to analog) converters, not A/D. I doubt seriously the typical human can hear the difference between 12bit and 16bit audio recording, especially when you consider that it needs to be converted from an analog voltage to mechanical movement by the speaker cone.

Whether the noise is analog or digital, what point is there in increasing the resolution if all you gain is more bits worth of noise?

[adas]

You're right about the D/A converter being in CD players. But there are switching too in the D/A converter just like A/D. Yet, my pocket mp3 player has a S/N ratio of 98dB (even tested by some reviewers). My only job is to find the finest headphones I can.

Keep in mind that Canon CMOS sensors aren't cheap at all. They're more than half the price of the entire camera. I bet they're very carefully designed then.

If the noise was the effect of the switching transistors at the quantizer level, the noise should've appeared more pattern like.
For example, in a completely defocused shot with a subject where the transition from black to white take place very slowly across the entire screen, we should see steeper light changes( wobbles) only at levels many 0's change to 1's or viceversa, all at once:

Ex:
0111 1111 1111 ---> 1000 0000 0000 strongest wobble (Level 2047 to 2048)
1011 0110 0111 ---> 1011 0110 1000 lighter wobble
0011 1011 1011 ---> 0011 1011 1100 lighter wobble

But this ain't happen.
Another noise type generated in the converter is the quantizer error, wich usually is limited to the least significant bit's uncertainity. Negligible.